Adaptive Red Panda Optimization for Feature Extraction in Diabetic Retinopathy Detection Using Deep Learning

Diabetic Retinopathy (DR) is a diabetes-related eye disease that affects the light-sensitive tissue of the retina and can lead to vision loss if not detected early. Traditional diagnostic approaches often overlook the value of ophthalmic imaging and are typically time-consuming and costly. In this study, we propose an Adaptive Red Panda Optimization-based Deep Convolutional Neural Network (ARPO-based DCNN) for effective DR detection. The methodology involves preprocessing retinal fundus images with a median filter, segmenting lesions using U-Net, and utilizing both the segmented and original images as input to a DCNN, which is trained with the ARPO algorithm-a combination of Red Panda Optimization (RPO) and adaptive mechanisms. For robust evaluation, we employed the publicly available Indian Diabetic Retinopathy image Dataset (IDRiD), which comprises high-resolution, annotated fundus images representing various stages and lesion types of DR, making it a standard benchmark in the field. Experimental results demonstrate that our ARPO-based DCNN achieves superior diagnostic performance, attaining an accuracy of 90.582%, sensitivity of 92.016%, and specificity of 90.272%, thereby highlighting its potential for reliable and automated DR screening